Location-based services (LBS) have gained increasing importance in our everyday lives and serve as the foundation for many smartphone applications. Whereas Global Navigation Satellite Systems (GNSS) enable reliable position estimation outdoors, there does not exist any comparable gold standard for indoor localization yet. Wireless local area network (WLAN) fingerprinting is still a promising and widely adopted approach to indoor localization, since it does not rely on preinstalled hardware but uses the existing WLAN infrastructure typically present in buildings. The accuracy of the method is, however, limited due to unstable fingerprints, etc. Deep learning has recently gained attention in the field of indoor localization and is also utilized to increase the performance of fingerprinting-based approaches. Current solutions can be grouped into models that either estimate the exact position of the user (regression) or classify the area (pre-segmented floor plan) or a reference location. We propose a model, DeepLocBox (DLB), that offers reliable area localization in multi-building/multi-floor environments without the prerequisite of a pre-segmented floor plan. Instead, the model predicts a bounding box that contains the user’s position while minimizing the required prediction space (size of the box). We compare the performance of DLB with the standard approach of neural network-based position estimation and demonstrate that DLB achieves a gain in success probability by 9.48% on a self-collected dataset at RWTH Aachen University, Germany; by 5.48% for a dataset provided by Tampere University of Technology (TUT), Finland; and by 3.71% for the UJIIndoorLoc dataset collected at Jaume I University (UJI) campus, Spain.

Pedestrian localization within large-scale multi-building/multi-floor indoor environments remains a challenging task. Fingerprinting-based approaches are particularly suited for such large-scale deployments due to their low requirements of hardware installments. Recently, the fingerprinting problem has been addressed by deep learning. Existing models are mostly task specific by providing floor classification or position estimation within a small area. A strategy to support localization within large-scale environments is to sequentially apply hierarchical models. This has several drawbacks including missing scalability and increased deployment complexity on smartphones. We propose a unifying approach based on training a single neural network that classifies the building/floor and predicts the position in a single forward-pass of the network. Our model classifies a grid cell and performs within grid cell regression, which solves the performance degradation of applying regression within large areas. To reduce the error in case of misclassified grid cells, we propose a novel technique called multi cell encoding learning (multi-CEL), where a model simultaneously learns several redundant position representations within an overlapping grid cell encoding. For three public WLAN fingerprinting datasets, we demonstrate that multi- CEL surpasses existing state-of-the-art multi-task learning neural networks and even outperforms regression neural networks explicitly trained for 2D-positioning by up to 17%.

The accuracy of fingerprinting-based indoor localization correlates with the quality and up-to-dateness of collected training data. Perpetual crowdsourced data collection reduces manual labeling effort and provides a fresh data base. However, the decentralized collection comes with the cost of heterogeneous data that causes performance degradation. In settings with imperfect data, area localization can provide higher positioning guarantees than exact position estimation. Existing area localization solutions employ a static segmentation into areas that is independent of the available training data. This approach is not applicable for crowdsoucred data collection, which features an unbalanced spatial training data distribution that evolves over time. A segmentation is required that utilizes the existing training data distribution and adapts once new data is accumulated. We propose an algorithm for data-aware floor plan segmentation and a selection metric that balances expressiveness (information gain) and performance (correctly classified examples) of area classifiers. We utilize supervised machine learning, in particular, deep learning, to train the area classifiers. We demonstrate how to regularly provide an area localization model that adapts its prediction space to the accumulating training data. The resulting models are shown to provide higher reliability compared to models that pinpoint the exact position.

Ground-source heat pumps with borehole heat exchangers (BHE) are an efficient and sustainable option to heat and cool buildings. The design and performance of BHEs strongly depend on the thermal conductivity of the subsurface. Thus, the first step in BHE planning is often assisted by a map representing the thermal conductivity of a region created from existing data. Such estimates have high uncertainty, which is rarely quantified. In addition, different methods for estimating thermal conductivity are used, for example, by the German federal states, resulting in incomparable estimates. To enable a consistent thermal conductivity estimation across state or country borders, we present a workflow for automatically estimating the thermal conductivity and its uncertainty up to user-defined BHE lengths. Two methods, which assess the thermal conductivity on different scales, are developed. Both methods are (1) based on subsurface data types which are publicly available as open-web services, and (2) account for thermal conductivity uncertainty by estimating its lowest, mean, and maximum values. The first method uses raster data, e.g., of surface geology and depth to groundwater table, and provides a large-scale estimate of the thermal conductivity, with high uncertainty. The second method improves the estimation for a small, user-defined target area by calculating the thermal conductivity based on the available borehole data in that area. The presented approach’s novelty is a web-based geodata infrastructure that seamlessly connects data provision and calculation processes, with a geoportal as its central user interface. To demonstrate the approach, we use data from the federal state of Hamburg and compare the results of two target areas with the thermal conductivity estimation by the Geological Survey of Hamburg. Depending on the selected region, differences between the two estimates can be considerable (up to 1.2 $\mathrm{W} {\mathrm{m}}^{-1} {\mathrm{K}}^{-1}$). The differences are primarily due to the selection of the thermal property database and the consideration of wet and dry rock. The results emphasize the importance of considering and communicating uncertainty in geothermal potential estimates.

Fingerprinting-based indoor localization is a cost-effective approach to provide coarse-grained indoor location information for pedestrian mass-market applications without the requirement of installing additional positioning infrastructure. While most solutions aim at pinpointing the exact location of a user, estimating a zone/area is a promising approach to achieve a more reliable prediction. Area localization predominantly utilizes a predetermined building model segmentation to obtain zone/area labels for collected fingerprints. We propose a novel approach to multifloor indoor area localization by directly predicting polygon zones that contain the position of the user. Our model learns to construct the zones from the wall segments and thus predicted areas have a high conformity to the underlying building model (semantic expressiveness). On a self-collected as well as on a public fingerprinting data set, we compare our model with two reference approaches. We demonstrate that the utilized surface areas of the polygons are on average up to 50% smaller than those of the reference models and provide a high semantic expressiveness without requiring manual floor plan segmentation.